(In reply to comment #9) > Sorry I'm out of ideas. Can the regression be reproduced by some existing > piglit test maybe? I'll run a piglit before and after the change... BTW you can't reproduce this with the http://pavel.ondracka.cz/EVE.trace trace?